You need to first calculate all your updates as if the wieghts weren't shared, but just store them, don't actually do any updating yet.
Let $w_k$ be some weight that appears at locations $I_k = \{(i,j) \colon w_{i,j} = w_k\}$ in your network and $\Delta w_{i,j} = -\eta \frac{\partial J}{\partial w_{i,j}} $ where $\eta$ is the learning rate and $J$ is your objective function. Note that at this point if you didn't have weight sharing you would just upade $w_{i,j}$ as
$$
w_{i,j} = w_{i,j} + \Delta w_{i,j}.
$$
To deal with the shared weights you need to sum up all the individual updates. So set
$$
\Delta w_k = \sum_{(i,j) \in I_k} \Delta w_{i,j}
$$
and then update
$$
w_k = w_k + \Delta w_k.
$$
I'll first try to share some intuition behind CNN and then comment the particular topics you listed.
The convolution and sub-sampling layers in a CNN are not different from the hidden layers in a common MLP, i. e. their function is to extract features from their input. These features are then given to the next hidden layer to extract still more complex features, or are directly given to a standard classifier to output the final prediction (usually a Softmax, but also SVM or any other can be used). In the context of image recognition, these features are images treats, like stroke patterns in the lower layers and object parts in the upper layers.
In natural images these features tend to be the same at all locations. Recognizing a certain stroke pattern in the middle of the images will be as useful as recognizing it close to the borders. So why don't we replicate the hidden layers and connect multiple copies of it in all regions of the input image, so the same features can be detected anywhere? It's exactly what a CNN does, but in a efficient way. After the replication (the "convolution" step) we add a sub-sample step, which can be implemented in many ways, but is nothing more than a sub-sample. In theory this step could be even removed, but in practice it's essential in order to allow the problem remain tractable.
Thus:
- Correct.
- As explained above, hidden layers of a CNN are feature extractors as in a regular MLP. The alternated convolution and sub-sampling steps are done during the training and classification, so they are not something done "before" the actual processing. I wouldn't call them "pre-processing", the same way the hidden layers of a MLP is not called so.
- Correct.
A good image which helps to understand the convolution is CNN page in the ULFDL tutorial. Think of a hidden layer with a single neuron which is trained to extract features from $3 \times 3$ patches. If we convolve this single learned feature over a $5 \times 5$ image, this process can be represented by the following gif:
![enter image description here](https://i.stack.imgur.com/I7DBr.gif)
In this example we were using a single neuron in our feature extraction layer, and we generated $9$ convolved features. If we had a larger number of units in the hidden layer, it would be clear why the sub-sampling step after this is required.
The subsequent convolution and sub-sampling steps are based in the same principle, but computed over features extracted in the previous layer, instead of the raw pixels of the original image.
Best Answer
When you are trying to recognize objects, edges, etc., the algorithms generally look for deviation from nearby or summed/averaged elements. The deviation could be negative or positive, but in most cases it is a "recognition" regardless of the polarity. The algorithms impart polarity due to the mathematics, but the interpretation doesn't care if it is a deviation up or down.
However, some filters and algorithms do carry information in the polarity. If you combine one where polarity matters with one where polarity does not, then you want to be sure you are adding a "recognition" properly, so the filter is augmented appropriately and not inappropriately pushed more negative or more positive.